At least now people will start to focus on how bloated and slow software has become in the meantime. And stop referring to it as “tech”.
Compute time is cheaper than man hours.
The Great CPU Stagnation
31–40 of 222 posts
Re: The Great CPU Stagnation
#32At least now people will start to focus on how bloated and slow software has become in the meantime. And stop referring to it as “tech”.
I (unfortunately) doubt it. I think the reality with our current situation is that only hobbyists and open source people care enough about that to do anything about it. Everyone else is doing Scrum so hard that they can't think of much outside the current sprint, and any "pre-mature optimization" is evil and must be avoided. The result will be more of the same bloat. Competition won't help because everybody is using…
Oh, I don't think it's quite that grim. When development teams can no longer easily spend their way out of their inefficient code, being able to write high performance code becomes a competitive advantage. I'd even go so far as to say that individual developers looking to get ahead in their careers should pay less attention to flashy tech fads and more at foundational stuff like how to diagnose performance problems and write efficient code.
Re: The Great CPU Stagnation
#33Re: The Great CPU Stagnation
#34IPC data seems a bit sus
It's the average IPC improvement from the AMD marketing slides - Zen 2 over Zen, 15%, Zen 3 over Zen 2, 19% and Zen 4 over Zen 3 at 13%, so Zen 3 over Zen is 37% and Zen 4 over Zen is 55% (maybe a typo in the article)
Re: The Great CPU Stagnation
#35We've known about this for a long time. Everyone expected it to happen. There are some key upcoming technologies that have the potential to cause a step in scaling (CFETs, backside power delivery) but it's still not going to be anywhere near Moore's law levels. I think this is part of why GPU power is skyrocketing and why Apple, Qualcomm, and the like are trying to shift towards services.
Well, and photonics, quantum computing, etc. But they're not there yet.
Re: The Great CPU Stagnation
#36This has to be coupled with 'The Great Software Decay'. I hope we will have 'real' innovation in Software Engineering where people take writing performant code seriously and many Software Engineers actually know what a compiler does and what the CPU does etc.
The frameworks have been helpful but at the same time -- rounding buttons is not software engineering.
Re: The Great CPU Stagnation
#37The answer to this might be more application specific accelerators and 3D stacking. You can't afford to have all of a chip switched on at once because of the power consumption and dissipation, so you build more optimised accelerators and keep flipping between them as you encounter different pieces of code that may benefit from each accelerator. Only a fraction of the device is ever in use at once. You 3D stack the ch…
It would be cool if chiplets got to the point where, at least, an OEM… someone Dell sized… could actually differentiate themselves by mix-and-matching a group of accelerators. Bringing back consumer-visible differentiation in CPUs (other than Apple against the world) would be nice for the market I think.
Re: The Great CPU Stagnation
#38IPC data seems a bit sus
It's the average IPC improvement from the AMD marketing slides - Zen 2 over Zen, 15%, Zen 3 over Zen 2, 19% and Zen 4 over Zen 3 at 13%, so Zen 3 over Zen is 37% and Zen 4 over Zen is 55% (maybe a typo in the article)
Zen 1 was terrible and is not a useful baseline. It was obsolete at launch. Nobody bought them except companies who were paid to take them and agree to be in a press release. It was literally half the speed of Skylake on server workloads.
Re: The Great CPU Stagnation
#39The cores/money comparison doesn't looks like it accounts for inflation.
Indeed. I was also wondering why people would buy newer chips if they don’t show performance increases per dollar.
Re: The Great CPU Stagnation
#40We've known about this for a long time. Everyone expected it to happen. There are some key upcoming technologies that have the potential to cause a step in scaling (CFETs, backside power delivery) but it's still not going to be anywhere near Moore's law levels. I think this is part of why GPU power is skyrocketing and why Apple, Qualcomm, and the like are trying to shift towards services.
Well the main problem is resistance isn't it? Most of the power "used" is to get electrons to flow fast enough for the logic gates to settle for a specific clock frequency and the resistive losses to heat. The only real way forward that isn't a temporary workaround seems finding a new type of semiconductor that has lower overall resistance than silicon. Whoever figures out how to dope graphene and produce wafers with…
When talking about resistance and materials, it's also important to note that silicon has relatively low optimal operating temperatures compared to some of the other semiconductors available. This limits the amount of voltage you can pump into it (because the resistance mean higher V leads to heat), and voltage correlates with clock frequency. GaN has already seen success in chargers, and silicon carbide is another promising material. We can't achieve the low level of defects needed for small process nodes yet, though.
Disclaimer: I'm not a material scientist, so this is probably only partly correct.